About this project

Methodology & Standards

coroner.ai is an experiment: point a deep-research AI at a prominent death and publish what it finds. It is not a verdict engine. It is a reading of the public record with the homework shown.

How a case file gets made

Each case runs through a fixed pipeline before anything is published:

1. Sweep. Several independent AI researchers each work one angle of the case — the official record and timeline, the forensic details as publicly reported, the alternative theories and how investigators answered them, and the surrounding context and aftermath. They work blind to each other, so one angle's framing can't contaminate another's.
2. Corroborate. Primary sources — police statements, medical-examiner statements, contemporaneous reporting, peer-reviewed research — outrank secondary summaries. A claim needs independent corroboration to be labeled confirmed. Where credible sources disagree, the disagreement is published, not silently resolved.
3. Break. A separate adversarial model attacks the merged findings before publication: it hunts for overreach (a "confirmed" resting on one source), defamation risk, internal contradictions, and citations that don't actually support their claims.
4. Publish with labels. What survives goes into the case file, each claim carrying a status label and its sources.

The labels

The labels are the product. If a label is wrong, the site is wrong.

confirmed — supported by multiple independent sources or by primary/official ones.

reported — published, but single-sourced or never officially confirmed (leaked figures live here).

disputed — credible sources conflict; both readings are shown.

unknown — sealed, never established, or only estimated.

Editorial standards

1. Official rulings are presented as what they are. We do not overrule medical examiners. Where the AI's reading concurs with or diverges from an official finding, it says so explicitly — with a confidence level and reasons.
2. Allegations are attributed. Always. Especially where they involve living people, an allegation appears with who made it and what investigators concluded about it. An allegation repeated without its rebuttal is just an accusation.
3. No invented citations. Every source on a case file was located by the research pipeline. Where a citation is still pending a research pass, the file says so instead of pretending.
4. Sealed is sealed. We never claim knowledge of documents that are not public — autopsy reports, sealed records. We tell you they're sealed instead.
5. Gaps get named. What could not be established is listed on the file, because an unstated gap reads as a verified fact.

Sensitive content

Many prominent deaths involve suicide. Case files record the facts plainly and skip gratuitous detail, following reporting-safety norms: no glamorization, no unnecessary method specifics, resources on every page. If you or someone you love is struggling, call or text 988 (US) or find international help at findahelpline.com.

Corrections

If a claim is mislabeled or a source doesn't hold up, we want to know. A corrections contact is coming with the public launch; corrected files note what changed and when.

Disclaimers

coroner.ai is not affiliated with any medical examiner, coroner, or law-enforcement agency. Case files are AI-generated analysis of public records and reporting; they can contain errors, and they are not legal, medical, or investigative advice. Official records belong to the offices that hold them.